EMPIRICAL RESULTS 1 Data and Sample Selection

The empirical data are collected from the TEJ database for the period from 2003 to 2006. Our preliminary sample consists of 1,228 firm-year observations. After subtracting observations with missing data and outliers,⁹ we obtain 1,053 observations for estimating the persistence of abnormal earnings and 756 observations for estimating the Ohlson model. Table 1 demonstrates the sample selection process.

[Insert Table 1 here]

4.2 Sample Distribution, Descriptive Statistics, and Correlations

Table 2 indicates that the numbers of firm-year observations are roughly the same across the period 2003~2006. However, there are considerable variations among sub-industries within the IT industry. For example, as shown on the last column of Table 2, 21.96% of the companies are manufacturers of electronic parts / components, followed by computer peripheral equipment (18.12%), optoelectronic (16.67%), and semiconductor (16.53%). Only 8.73%, 6.22%, and 1.19% of the companies belong to Internet communication, electronic product distribution, and information service, respectively. Finally, some of the largest companies are classified as “Others” because of their high diversification (e.g., Foxconn Technology Group, the world largest provider of CEM, EMS, ODM and CMMS).¹⁰

and return on total assets, with a weight of 1/9 assigned to each ratio), liquidity and solvency (including quick ratio, interest payment ratio, and loan dependence, with a weight of 1/9 assigned to each ratio), business activity (including average collection period for receivables and average inventory selling days, with a weight of 1/18 assigned to each ratio), and size (including total revenue and total assets, with a weight of 1/9 assigned to each ratio).

9To control for outliers, we trim observations that fall outside the upper and lower 1% of the empirical distributions for both the dependent and independent variables.

10Contract electronic manufacturers (CEM) are companies that offer contracts for electronic assembly for other original equipment manufacturers (OEM). Generally, a CEM does not post its brand name on any product, and both the design and the brand name belongs to the OEM. In contrast, electronics manufacturing service (EMS) providers are companies that design, test, manufacture, distribute, and provide repair services for electronic components and assemblies for OEM. Original design manufacturers (ODM) are companies that manufacture products which ultimately will be branded by another company for sale. ODM companies allow the brand firm to produce without having to engage in the organization or running of a factory. Finally, component module move service (CMMS) providers are companies offering joint development manufacturing (JDVM) and joint design manufacturing

12

[Insert Table 2 here]

Table 3 summarizes the definitions of all variables. Table 4 presents the descriptive statistics of these variables. All monetary amounts are measured by New Taiwan (NT) dollars (with an exchange rate around US$1 = NT$32). As reported in Table 4, the year-end stock price has a mean value of NT$27.421, ranging from NT$7.00 to NT$146.42. In addition, the average book value per share and abnormal earnings per share are NT$17.867 and NT$0.179, respectively. The mean value of analysts’ forecasts is NT$0.881. Notably, the foreign institutional investors’ ownership percentages display considerable dispersion, with a minimum of zero to a maximum of 73.1%. Our Ohlson model estimations thus control for the possible effect of foreign institution ownership on the value relevance of intellectual capital.

[Insert Tables 3 and 4 here]

Table 4 also shows substantial variations between and within four types of intellectual capital. For example, the mean values of EDU and BONUS are 0.115 (ranging from 0 to 0.679) and 0.150 (ranging from 0 to 0.769), respectively, indicating that Taiwan IT companies’ human capital is more likely to be driven by paying more bonuses to their employees. In contrast, the mean values of RD and ROYALTY are 0.037 (ranging from 0 to 0.195) and 0.002 (ranging from 0 to 0.115), respectively, implying that sample firms tend to develop innovation capital mainly through their own research activities rather than acquiring patents from outside parties. The mean values of ADMIN and TURNOVER are 0.029 (ranging from 0 to 0.126) and 0.010 (ranging from 0.002 to 0.037), respectively, suggesting that Taiwan’s IT companies invest more administrative expenses to manage their process capital. Finally, the mean values of PROMOTION and CREDIT are 0.039 (ranging from zero to 0.211) and 3.413 (with Q2 = 2, median = 3, and Q3 = 4), respectively, indicating that most of these IT companies have fairly good credit ratings based on which they build up their relational capital. To the extent that the above eight variables capture the capacity of their corresponding intellectual capital dimensions, Table 4 implies that Taiwan’s IT companies have a propensity to focus more on the human capital than on the other three capital types.

Table 5 reports the Pearson and Spearman correlations for the variables. Consistent with Dechow et al.

(1999) and Yu et al. (2003), Table 5 demonstrates that firms’ stock prices are positively correlated with their book values (BV), abnormal earnings (X ^a), analysts’ forecasts (V ^f), and shareholders' deductible tax amount

(JDSM). Since a CMMS takes the advantages of CEM and ODM, it can effectively reduce its production costs and speedup the production process. The CMMS model was initially developed and adopted by Foxconn since 1998.

13

(CT). Also, stock prices are significantly correlated with most of the intellectual capital variables (e.g., EDU, BONUS, ADMIN, TURNOVER, PROMOTION, and CREDIT) in the predicted directions, suggesting that investors seem to be able to detect and incorporate this intellectual capital information into their business valuation process. Two things are worth noting. First, many of the correlation coefficients are significant among independent variables. Therefore, multicollinearity may adversely affect our statistical testing power.

We will use the variance inflation factor (VIF) to test if our empirical results are subjected to this multicollinearity problem. Second, untabulated statistics indicate that the Pearson and Spearman correlation coefficients between the basic ranks (which is labeled CREDIT in Table 5) and TEJ’s TCRI are 0.7274 and 0.7494, respectively. Both coefficients are significant at the 1% significance level. Therefore, our use of the basic ranks provides an appropriate proxy for firms’ real TCRI.

[Insert Table 5 here]

4.3 Determinants of the Persistence of Abnormal Earnings

Table 6 presents the empirical result for the determinants of the persistence of abnormal earnings, with all parameter estimates significant at the two-tailed 1% level. Except for ω₂, persistence is hypothesized to decrease when earnings contain more transitory accounting items, which are measured by the empirical constructs q2 and q3. The significance of ω₃ and ω₄ indicates that the less the nonrecurring items and accounting accruals, the more the abnormal earnings in the next period. In contrast, the significance of ω₅ implies that an increase in cash dividend would result in an increase in next period’s abnormal earnings.

[Insert Table 6 here]

The significantly positive coefficient ω₂, which is inconsistent with our prediction, suggests that Taiwan’s IT companies having high abnormal accounting rate of returns do not exhibit mean-reverting on their accounting rate of returns. In fact, these companies enjoy an even higher increase in their abnormal earnings mainly due to a substantial increase in sales growth. For example, Foxconn Technology Group’s net sales increase by 28.7%, 59.7% and 34.7% in 2004, 2005, and 2006, respectively. Likewise, Quanta Computer’s (the world’s largest notebook ODM / OEM company) operating sales increase by 11.0%, 24.2% and 14.5%

in the same period. Surprisingly, the ASUSTek Computer (the world’s largest motherboard manufacturer who introduced the EeePC in 2007) has a more than 20-fold increase in sales from 4.9% in 2004 to 130.3%

in 2005 and 114.8% in 2006. These and many other IT companies have some important characteristics in

14

common. First, they successfully establish their brand names by continuously designing high-end electronic products that lead the global market. Second, they develop more efficient manufacturing processes to cut down the costs, reduce the defective rate, shorten the production cycle, and adapt to technological changes faster. Finally, they maintain a close networked relation with their global vendors and customers. Because of these salient features, together with the government’s long-term financial supports, Taiwan’s IT companies are able to sustain a considerable sales increase in consecutive years, leading to a positive association between their abnormal accounting rates of return and the persistence of abnormal earnings. Our study thus contributes to the literature by providing empirical evidence against prior studies’ finding that companies having extreme accounting rate of returns exhibit stronger mean-reverting on their accounting rate of returns (e.g., Dechow et al. 1999; Freeman et al. 1982).

4.4 Value Relevance of Intellectual Capital

We first run Model (1) without considering the intellectual capital variables. As depicted in the first and second columns of Table 7, all the independent variables are significant in their predicted directions (two-tailed p < 0.01). A fairly high adjusted R² of 0.5273 implies that the Ohlson model is appropriate for Taiwan’s stock market (F = 141.55, p < 0.000).

[Insert Table 7 here]

After taking intellectual capital variables into account, the third and fourth columns of Table 7 indicate several major findings. First, except for ROYALTY and TURNOVER (which are significant at the one-tailed 10% level), all other intellectual capital variables are significant at least at the two-tailed 5% level. This result, together with a 0.1008 (0.6281 – 0.5273) increase in adjusted R², suggests that the public disclosure of these eight intellectual capital variables is relevant to market participants in determining stock prices. In this regard, we provide empirical evidence showing that Taiwan’s stock market is able to incorporate IT companies’ various types of intellectual capital into the determination of equity values simultaneously.

Second, the relatively weaker significance of the innovation capital may arise from IT companies’

unwillingness to disclose more innovation-related information that could be beneficial to their potential competitors (García-Meca and Martínez 2007; Marston 1996). Another possible reason is that market participants place less emphasis on IT companies’ innovation activities because the outcomes of these activities are usually uncertain and unpredictable (Barker 1999; García-Meca and Martínez 2007).

15

Third, the coefficients of ROYALTY (i.e., a₁₀) and PROMOTION (i.e., a₁₃) are significant but not in the predicted direction. A significantly negative a₁₀ may capture the stock market’s view that higher royalty payment ratio signals weaker R&D ability. To maintain leadership in the highly competitive global IT market, patents and copyrights developed by the companies are deemed crucial for their long-term survival.

In contrast, the significantly negative coefficient a13 may reflect the fact that, since many of the buyers of Taiwan’s IT (especially semiconductor and electronic components) companies are manufacturers rather than individual customers, advertising expenses may create little value to the relational capital but simply increase marketing expenses. This finding provides management implication that advertisement would not be a value-creation vehicle to the IT companies.

Finally, Table 7 shows that abnormal earnings X ^a becomes insignificant when all intellectual capital variables are included in Model (1). Because X ^a captures the “unexpected” part of the realized earnings, the vanishing of its significance suggests that intellectual capital accounts for a large portion of the unexpected earnings. This finding contributes to the literature by emphasizing the importance of considering intellectual capital when one uses the Ohlson model to examine value relevance and business valuation issues. Note that the VIFs of all independent variables reported in Table 7 are smaller than 2.27, implying that our empirical results are not subjected to multicollinearity (Kleinbaum et al. 1997).

4.5 Is the Use of a Constant Discount Rate Appropriate?

We argue that the use of a constant discount rate r could be theoretically and empirically inappropriate because it explicitly assumes that all firms have the same rate of return on equity across multiple periods, resulting in biased estimation of abnormal earnings. To elaborate this issue, we run Model (1) using a constant r manipulated at five levels: 10%, 11%, 12%, 13%, and 14%. Table 8 displays that, across these five r levels, the coefficient of abnormal earnings X^a is significant at two-tailed 1% level while the coefficients of ROYALTY (an innovation capital) and TURNOVER (a process capital) become insignificant.

These results contrast with Table 7, which shows that the explanatory power of abnormal earnings has been replaced by intellectual capital.

[Insert Table 8 here]

A comparison of Tables 7 and 8 reveals two empirical implications. First, the use of constant r’s seems to mask, dilute, or even distort the value relevance of intellectual capital during the business valuation

16

process. A notable example is CREDIT, whose coefficient remains significant (two-tailed p < 0.01) but change from negative to positive. Because higher TCRI rank represents higher credit risk, a positive coefficient of CREDIT is inconsistent with recent empirical finding that downgrade credit rating is associated with decreases in stock prices (e.g., Parnes 2008). Second, we conjecture that the significance of the abnormal earnings and the fairly high adjusted R² (which increases by almost 8% in Table 8) documented in prior studies using the Ohlson model could be the result of assuming a constant r corss-sectionally and intertemporally. In light of these, a suggested refinement of future studies would be the use of different rates of return on equity that accommodate variations across firms and years.

4.6 Robustness Tests:

4.6.1 Use April-end stock prices

Since the intellectual capital information will not be publicly available until the release of the annual reports (whose official deadline for all Taiwanese listed companies is April 30), we also use firm i’s stock prices at the end of April in year t+1 as the dependent variable. The empirical results are similar to those reported in Table 7. Therefore, our empirical findings do not change under different stock price time frames.

4.6.2 Use the last consensus earnings forecast per share

Barron et al. (2002) finds that analyst consensus is lower for high-tech manufacturing companies due to their high uncertainty in future earnings associated with intangible assets. Because prior studies have documented that individual analysts’ forecasts will be more accurate as firms’ earnings announcements approach (e.g., Francis and Philbrick 1993; Lim 2001; O'Brien, 1988), it is possible that the level of analyst consensus will be higher for forecasts made near the year-end than those made early in the year. We thus measure V ^f by the last consensus earnings forecast per share at the end of year t to re-estimate model (1).

The empirical results are similar to those reported in Table 7.

4.6.3 Alternative measure of shareholders’ imputed credits

Because Taiwan’s GAAP requires that companies disclose shareholders’ taxable dividend balances in the footnotes, market participants may interpret these numbers mechanically without taking into account the accrued tax payables to be paid in the next year. To test whether the results reported in Table 7 are subjected to market participants’ “functional fixation,” we exclude accrued tax payables from CT and re-estimate model (1). Our conclusions remain the same under this alternative CT measure.

17 4.6.4 Eliminate Y2004

Because Taiwan’s 2004 Presidential Election is a political event that has never occurred before in Taiwan’s history, the above analyses may be subjected to the uniqueness of this non-economic event, leading to weak generalization of our empirical results. We eliminate Y2004 and re-estimate model (1). The results obtained in Table 7 remain unchanged.

5. CONCLUDING REMARKS